1 #include <linux/etherdevice.h> 2 #include <linux/if_macvlan.h> 3 #include <linux/if_vlan.h> 4 #include <linux/interrupt.h> 5 #include <linux/nsproxy.h> 6 #include <linux/compat.h> 7 #include <linux/if_tun.h> 8 #include <linux/module.h> 9 #include <linux/skbuff.h> 10 #include <linux/cache.h> 11 #include <linux/sched.h> 12 #include <linux/types.h> 13 #include <linux/slab.h> 14 #include <linux/wait.h> 15 #include <linux/cdev.h> 16 #include <linux/idr.h> 17 #include <linux/fs.h> 18 #include <linux/uio.h> 19 20 #include <net/net_namespace.h> 21 #include <net/rtnetlink.h> 22 #include <net/sock.h> 23 #include <linux/virtio_net.h> 24 #include <linux/skb_array.h> 25 26 /* 27 * A macvtap queue is the central object of this driver, it connects 28 * an open character device to a macvlan interface. There can be 29 * multiple queues on one interface, which map back to queues 30 * implemented in hardware on the underlying device. 31 * 32 * macvtap_proto is used to allocate queues through the sock allocation 33 * mechanism. 34 * 35 */ 36 struct macvtap_queue { 37 struct sock sk; 38 struct socket sock; 39 struct socket_wq wq; 40 int vnet_hdr_sz; 41 struct macvlan_dev __rcu *vlan; 42 struct file *file; 43 unsigned int flags; 44 u16 queue_index; 45 bool enabled; 46 struct list_head next; 47 struct skb_array skb_array; 48 }; 49 50 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) 51 52 #define MACVTAP_VNET_LE 0x80000000 53 #define MACVTAP_VNET_BE 0x40000000 54 55 #ifdef CONFIG_TUN_VNET_CROSS_LE 56 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 57 { 58 return q->flags & MACVTAP_VNET_BE ? false : 59 virtio_legacy_is_little_endian(); 60 } 61 62 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp) 63 { 64 int s = !!(q->flags & MACVTAP_VNET_BE); 65 66 if (put_user(s, sp)) 67 return -EFAULT; 68 69 return 0; 70 } 71 72 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp) 73 { 74 int s; 75 76 if (get_user(s, sp)) 77 return -EFAULT; 78 79 if (s) 80 q->flags |= MACVTAP_VNET_BE; 81 else 82 q->flags &= ~MACVTAP_VNET_BE; 83 84 return 0; 85 } 86 #else 87 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 88 { 89 return virtio_legacy_is_little_endian(); 90 } 91 92 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp) 93 { 94 return -EINVAL; 95 } 96 97 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp) 98 { 99 return -EINVAL; 100 } 101 #endif /* CONFIG_TUN_VNET_CROSS_LE */ 102 103 static inline bool macvtap_is_little_endian(struct macvtap_queue *q) 104 { 105 return q->flags & MACVTAP_VNET_LE || 106 macvtap_legacy_is_little_endian(q); 107 } 108 109 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) 110 { 111 return __virtio16_to_cpu(macvtap_is_little_endian(q), val); 112 } 113 114 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) 115 { 116 return __cpu_to_virtio16(macvtap_is_little_endian(q), val); 117 } 118 119 static struct proto macvtap_proto = { 120 .name = "macvtap", 121 .owner = THIS_MODULE, 122 .obj_size = sizeof (struct macvtap_queue), 123 }; 124 125 /* 126 * Variables for dealing with macvtaps device numbers. 127 */ 128 static dev_t macvtap_major; 129 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 130 static DEFINE_MUTEX(minor_lock); 131 static DEFINE_IDR(minor_idr); 132 133 #define GOODCOPY_LEN 128 134 static const void *macvtap_net_namespace(struct device *d) 135 { 136 struct net_device *dev = to_net_dev(d->parent); 137 return dev_net(dev); 138 } 139 140 static struct class macvtap_class = { 141 .name = "macvtap", 142 .owner = THIS_MODULE, 143 .ns_type = &net_ns_type_operations, 144 .namespace = macvtap_net_namespace, 145 }; 146 static struct cdev macvtap_cdev; 147 148 static const struct proto_ops macvtap_socket_ops; 149 150 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 151 NETIF_F_TSO6 | NETIF_F_UFO) 152 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 153 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 154 155 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 156 { 157 return rcu_dereference(dev->rx_handler_data); 158 } 159 160 /* 161 * RCU usage: 162 * The macvtap_queue and the macvlan_dev are loosely coupled, the 163 * pointers from one to the other can only be read while rcu_read_lock 164 * or rtnl is held. 165 * 166 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 167 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 168 * q->vlan becomes inaccessible. When the files gets closed, 169 * macvtap_get_queue() fails. 170 * 171 * There may still be references to the struct sock inside of the 172 * queue from outbound SKBs, but these never reference back to the 173 * file or the dev. The data structure is freed through __sk_free 174 * when both our references and any pending SKBs are gone. 175 */ 176 177 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 178 struct macvtap_queue *q) 179 { 180 struct macvlan_dev *vlan = netdev_priv(dev); 181 int err = -EINVAL; 182 183 ASSERT_RTNL(); 184 185 if (q->enabled) 186 goto out; 187 188 err = 0; 189 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 190 q->queue_index = vlan->numvtaps; 191 q->enabled = true; 192 193 vlan->numvtaps++; 194 out: 195 return err; 196 } 197 198 /* Requires RTNL */ 199 static int macvtap_set_queue(struct net_device *dev, struct file *file, 200 struct macvtap_queue *q) 201 { 202 struct macvlan_dev *vlan = netdev_priv(dev); 203 204 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 205 return -EBUSY; 206 207 rcu_assign_pointer(q->vlan, vlan); 208 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 209 sock_hold(&q->sk); 210 211 q->file = file; 212 q->queue_index = vlan->numvtaps; 213 q->enabled = true; 214 file->private_data = q; 215 list_add_tail(&q->next, &vlan->queue_list); 216 217 vlan->numvtaps++; 218 vlan->numqueues++; 219 220 return 0; 221 } 222 223 static int macvtap_disable_queue(struct macvtap_queue *q) 224 { 225 struct macvlan_dev *vlan; 226 struct macvtap_queue *nq; 227 228 ASSERT_RTNL(); 229 if (!q->enabled) 230 return -EINVAL; 231 232 vlan = rtnl_dereference(q->vlan); 233 234 if (vlan) { 235 int index = q->queue_index; 236 BUG_ON(index >= vlan->numvtaps); 237 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 238 nq->queue_index = index; 239 240 rcu_assign_pointer(vlan->taps[index], nq); 241 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 242 q->enabled = false; 243 244 vlan->numvtaps--; 245 } 246 247 return 0; 248 } 249 250 /* 251 * The file owning the queue got closed, give up both 252 * the reference that the files holds as well as the 253 * one from the macvlan_dev if that still exists. 254 * 255 * Using the spinlock makes sure that we don't get 256 * to the queue again after destroying it. 257 */ 258 static void macvtap_put_queue(struct macvtap_queue *q) 259 { 260 struct macvlan_dev *vlan; 261 262 rtnl_lock(); 263 vlan = rtnl_dereference(q->vlan); 264 265 if (vlan) { 266 if (q->enabled) 267 BUG_ON(macvtap_disable_queue(q)); 268 269 vlan->numqueues--; 270 RCU_INIT_POINTER(q->vlan, NULL); 271 sock_put(&q->sk); 272 list_del_init(&q->next); 273 } 274 275 rtnl_unlock(); 276 277 synchronize_rcu(); 278 sock_put(&q->sk); 279 } 280 281 /* 282 * Select a queue based on the rxq of the device on which this packet 283 * arrived. If the incoming device is not mq, calculate a flow hash 284 * to select a queue. If all fails, find the first available queue. 285 * Cache vlan->numvtaps since it can become zero during the execution 286 * of this function. 287 */ 288 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 289 struct sk_buff *skb) 290 { 291 struct macvlan_dev *vlan = netdev_priv(dev); 292 struct macvtap_queue *tap = NULL; 293 /* Access to taps array is protected by rcu, but access to numvtaps 294 * isn't. Below we use it to lookup a queue, but treat it as a hint 295 * and validate that the result isn't NULL - in case we are 296 * racing against queue removal. 297 */ 298 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 299 __u32 rxq; 300 301 if (!numvtaps) 302 goto out; 303 304 if (numvtaps == 1) 305 goto single; 306 307 /* Check if we can use flow to select a queue */ 308 rxq = skb_get_hash(skb); 309 if (rxq) { 310 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 311 goto out; 312 } 313 314 if (likely(skb_rx_queue_recorded(skb))) { 315 rxq = skb_get_rx_queue(skb); 316 317 while (unlikely(rxq >= numvtaps)) 318 rxq -= numvtaps; 319 320 tap = rcu_dereference(vlan->taps[rxq]); 321 goto out; 322 } 323 324 single: 325 tap = rcu_dereference(vlan->taps[0]); 326 out: 327 return tap; 328 } 329 330 /* 331 * The net_device is going away, give up the reference 332 * that it holds on all queues and safely set the pointer 333 * from the queues to NULL. 334 */ 335 static void macvtap_del_queues(struct net_device *dev) 336 { 337 struct macvlan_dev *vlan = netdev_priv(dev); 338 struct macvtap_queue *q, *tmp; 339 340 ASSERT_RTNL(); 341 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 342 list_del_init(&q->next); 343 RCU_INIT_POINTER(q->vlan, NULL); 344 if (q->enabled) 345 vlan->numvtaps--; 346 vlan->numqueues--; 347 sock_put(&q->sk); 348 } 349 BUG_ON(vlan->numvtaps); 350 BUG_ON(vlan->numqueues); 351 /* guarantee that any future macvtap_set_queue will fail */ 352 vlan->numvtaps = MAX_MACVTAP_QUEUES; 353 } 354 355 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 356 { 357 struct sk_buff *skb = *pskb; 358 struct net_device *dev = skb->dev; 359 struct macvlan_dev *vlan; 360 struct macvtap_queue *q; 361 netdev_features_t features = TAP_FEATURES; 362 363 vlan = macvtap_get_vlan_rcu(dev); 364 if (!vlan) 365 return RX_HANDLER_PASS; 366 367 q = macvtap_get_queue(dev, skb); 368 if (!q) 369 return RX_HANDLER_PASS; 370 371 if (__skb_array_full(&q->skb_array)) 372 goto drop; 373 374 skb_push(skb, ETH_HLEN); 375 376 /* Apply the forward feature mask so that we perform segmentation 377 * according to users wishes. This only works if VNET_HDR is 378 * enabled. 379 */ 380 if (q->flags & IFF_VNET_HDR) 381 features |= vlan->tap_features; 382 if (netif_needs_gso(skb, features)) { 383 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 384 385 if (IS_ERR(segs)) 386 goto drop; 387 388 if (!segs) { 389 if (skb_array_produce(&q->skb_array, skb)) 390 goto drop; 391 goto wake_up; 392 } 393 394 consume_skb(skb); 395 while (segs) { 396 struct sk_buff *nskb = segs->next; 397 398 segs->next = NULL; 399 if (skb_array_produce(&q->skb_array, segs)) { 400 kfree_skb(segs); 401 kfree_skb_list(nskb); 402 break; 403 } 404 segs = nskb; 405 } 406 } else { 407 /* If we receive a partial checksum and the tap side 408 * doesn't support checksum offload, compute the checksum. 409 * Note: it doesn't matter which checksum feature to 410 * check, we either support them all or none. 411 */ 412 if (skb->ip_summed == CHECKSUM_PARTIAL && 413 !(features & NETIF_F_CSUM_MASK) && 414 skb_checksum_help(skb)) 415 goto drop; 416 if (skb_array_produce(&q->skb_array, skb)) 417 goto drop; 418 } 419 420 wake_up: 421 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 422 return RX_HANDLER_CONSUMED; 423 424 drop: 425 /* Count errors/drops only here, thus don't care about args. */ 426 macvlan_count_rx(vlan, 0, 0, 0); 427 kfree_skb(skb); 428 return RX_HANDLER_CONSUMED; 429 } 430 431 static int macvtap_get_minor(struct macvlan_dev *vlan) 432 { 433 int retval = -ENOMEM; 434 435 mutex_lock(&minor_lock); 436 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 437 if (retval >= 0) { 438 vlan->minor = retval; 439 } else if (retval == -ENOSPC) { 440 netdev_err(vlan->dev, "Too many macvtap devices\n"); 441 retval = -EINVAL; 442 } 443 mutex_unlock(&minor_lock); 444 return retval < 0 ? retval : 0; 445 } 446 447 static void macvtap_free_minor(struct macvlan_dev *vlan) 448 { 449 mutex_lock(&minor_lock); 450 if (vlan->minor) { 451 idr_remove(&minor_idr, vlan->minor); 452 vlan->minor = 0; 453 } 454 mutex_unlock(&minor_lock); 455 } 456 457 static struct net_device *dev_get_by_macvtap_minor(int minor) 458 { 459 struct net_device *dev = NULL; 460 struct macvlan_dev *vlan; 461 462 mutex_lock(&minor_lock); 463 vlan = idr_find(&minor_idr, minor); 464 if (vlan) { 465 dev = vlan->dev; 466 dev_hold(dev); 467 } 468 mutex_unlock(&minor_lock); 469 return dev; 470 } 471 472 static int macvtap_newlink(struct net *src_net, 473 struct net_device *dev, 474 struct nlattr *tb[], 475 struct nlattr *data[]) 476 { 477 struct macvlan_dev *vlan = netdev_priv(dev); 478 int err; 479 480 INIT_LIST_HEAD(&vlan->queue_list); 481 482 /* Since macvlan supports all offloads by default, make 483 * tap support all offloads also. 484 */ 485 vlan->tap_features = TUN_OFFLOADS; 486 487 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 488 if (err) 489 return err; 490 491 /* Don't put anything that may fail after macvlan_common_newlink 492 * because we can't undo what it does. 493 */ 494 err = macvlan_common_newlink(src_net, dev, tb, data); 495 if (err) { 496 netdev_rx_handler_unregister(dev); 497 return err; 498 } 499 500 return 0; 501 } 502 503 static void macvtap_dellink(struct net_device *dev, 504 struct list_head *head) 505 { 506 netdev_rx_handler_unregister(dev); 507 macvtap_del_queues(dev); 508 macvlan_dellink(dev, head); 509 } 510 511 static void macvtap_setup(struct net_device *dev) 512 { 513 macvlan_common_setup(dev); 514 dev->tx_queue_len = TUN_READQ_SIZE; 515 } 516 517 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 518 .kind = "macvtap", 519 .setup = macvtap_setup, 520 .newlink = macvtap_newlink, 521 .dellink = macvtap_dellink, 522 }; 523 524 525 static void macvtap_sock_write_space(struct sock *sk) 526 { 527 wait_queue_head_t *wqueue; 528 529 if (!sock_writeable(sk) || 530 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 531 return; 532 533 wqueue = sk_sleep(sk); 534 if (wqueue && waitqueue_active(wqueue)) 535 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 536 } 537 538 static void macvtap_sock_destruct(struct sock *sk) 539 { 540 struct macvtap_queue *q = container_of(sk, struct macvtap_queue, sk); 541 542 skb_array_cleanup(&q->skb_array); 543 } 544 545 static int macvtap_open(struct inode *inode, struct file *file) 546 { 547 struct net *net = current->nsproxy->net_ns; 548 struct net_device *dev; 549 struct macvtap_queue *q; 550 int err = -ENODEV; 551 552 rtnl_lock(); 553 dev = dev_get_by_macvtap_minor(iminor(inode)); 554 if (!dev) 555 goto err; 556 557 err = -ENOMEM; 558 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 559 &macvtap_proto, 0); 560 if (!q) 561 goto err; 562 563 RCU_INIT_POINTER(q->sock.wq, &q->wq); 564 init_waitqueue_head(&q->wq.wait); 565 q->sock.type = SOCK_RAW; 566 q->sock.state = SS_CONNECTED; 567 q->sock.file = file; 568 q->sock.ops = &macvtap_socket_ops; 569 sock_init_data(&q->sock, &q->sk); 570 q->sk.sk_write_space = macvtap_sock_write_space; 571 q->sk.sk_destruct = macvtap_sock_destruct; 572 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 573 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 574 575 /* 576 * so far only KVM virtio_net uses macvtap, enable zero copy between 577 * guest kernel and host kernel when lower device supports zerocopy 578 * 579 * The macvlan supports zerocopy iff the lower device supports zero 580 * copy so we don't have to look at the lower device directly. 581 */ 582 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 583 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 584 585 err = -ENOMEM; 586 if (skb_array_init(&q->skb_array, dev->tx_queue_len, GFP_KERNEL)) 587 goto err_array; 588 589 err = macvtap_set_queue(dev, file, q); 590 if (err) 591 goto err_queue; 592 593 dev_put(dev); 594 595 rtnl_unlock(); 596 return err; 597 598 err_queue: 599 skb_array_cleanup(&q->skb_array); 600 err_array: 601 sock_put(&q->sk); 602 err: 603 if (dev) 604 dev_put(dev); 605 606 rtnl_unlock(); 607 return err; 608 } 609 610 static int macvtap_release(struct inode *inode, struct file *file) 611 { 612 struct macvtap_queue *q = file->private_data; 613 macvtap_put_queue(q); 614 return 0; 615 } 616 617 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 618 { 619 struct macvtap_queue *q = file->private_data; 620 unsigned int mask = POLLERR; 621 622 if (!q) 623 goto out; 624 625 mask = 0; 626 poll_wait(file, &q->wq.wait, wait); 627 628 if (!skb_array_empty(&q->skb_array)) 629 mask |= POLLIN | POLLRDNORM; 630 631 if (sock_writeable(&q->sk) || 632 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 633 sock_writeable(&q->sk))) 634 mask |= POLLOUT | POLLWRNORM; 635 636 out: 637 return mask; 638 } 639 640 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 641 size_t len, size_t linear, 642 int noblock, int *err) 643 { 644 struct sk_buff *skb; 645 646 /* Under a page? Don't bother with paged skb. */ 647 if (prepad + len < PAGE_SIZE || !linear) 648 linear = len; 649 650 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 651 err, 0); 652 if (!skb) 653 return NULL; 654 655 skb_reserve(skb, prepad); 656 skb_put(skb, linear); 657 skb->data_len = len - linear; 658 skb->len += len - linear; 659 660 return skb; 661 } 662 663 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 664 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 665 666 /* Get packet from user space buffer */ 667 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 668 struct iov_iter *from, int noblock) 669 { 670 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 671 struct sk_buff *skb; 672 struct macvlan_dev *vlan; 673 unsigned long total_len = iov_iter_count(from); 674 unsigned long len = total_len; 675 int err; 676 struct virtio_net_hdr vnet_hdr = { 0 }; 677 int vnet_hdr_len = 0; 678 int copylen = 0; 679 int depth; 680 bool zerocopy = false; 681 size_t linear; 682 683 if (q->flags & IFF_VNET_HDR) { 684 vnet_hdr_len = q->vnet_hdr_sz; 685 686 err = -EINVAL; 687 if (len < vnet_hdr_len) 688 goto err; 689 len -= vnet_hdr_len; 690 691 err = -EFAULT; 692 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from)) 693 goto err; 694 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 695 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 696 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 697 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 698 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 699 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 700 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 701 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 702 err = -EINVAL; 703 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 704 goto err; 705 } 706 707 err = -EINVAL; 708 if (unlikely(len < ETH_HLEN)) 709 goto err; 710 711 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 712 struct iov_iter i; 713 714 copylen = vnet_hdr.hdr_len ? 715 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 716 if (copylen > good_linear) 717 copylen = good_linear; 718 else if (copylen < ETH_HLEN) 719 copylen = ETH_HLEN; 720 linear = copylen; 721 i = *from; 722 iov_iter_advance(&i, copylen); 723 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 724 zerocopy = true; 725 } 726 727 if (!zerocopy) { 728 copylen = len; 729 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 730 if (linear > good_linear) 731 linear = good_linear; 732 else if (linear < ETH_HLEN) 733 linear = ETH_HLEN; 734 } 735 736 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 737 linear, noblock, &err); 738 if (!skb) 739 goto err; 740 741 if (zerocopy) 742 err = zerocopy_sg_from_iter(skb, from); 743 else 744 err = skb_copy_datagram_from_iter(skb, 0, from, len); 745 746 if (err) 747 goto err_kfree; 748 749 skb_set_network_header(skb, ETH_HLEN); 750 skb_reset_mac_header(skb); 751 skb->protocol = eth_hdr(skb)->h_proto; 752 753 if (vnet_hdr_len) { 754 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, 755 macvtap_is_little_endian(q)); 756 if (err) 757 goto err_kfree; 758 } 759 760 skb_probe_transport_header(skb, ETH_HLEN); 761 762 /* Move network header to the right position for VLAN tagged packets */ 763 if ((skb->protocol == htons(ETH_P_8021Q) || 764 skb->protocol == htons(ETH_P_8021AD)) && 765 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) 766 skb_set_network_header(skb, depth); 767 768 rcu_read_lock(); 769 vlan = rcu_dereference(q->vlan); 770 /* copy skb_ubuf_info for callback when skb has no error */ 771 if (zerocopy) { 772 skb_shinfo(skb)->destructor_arg = m->msg_control; 773 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 774 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 775 } else if (m && m->msg_control) { 776 struct ubuf_info *uarg = m->msg_control; 777 uarg->callback(uarg, false); 778 } 779 780 if (vlan) { 781 skb->dev = vlan->dev; 782 dev_queue_xmit(skb); 783 } else { 784 kfree_skb(skb); 785 } 786 rcu_read_unlock(); 787 788 return total_len; 789 790 err_kfree: 791 kfree_skb(skb); 792 793 err: 794 rcu_read_lock(); 795 vlan = rcu_dereference(q->vlan); 796 if (vlan) 797 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 798 rcu_read_unlock(); 799 800 return err; 801 } 802 803 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 804 { 805 struct file *file = iocb->ki_filp; 806 struct macvtap_queue *q = file->private_data; 807 808 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 809 } 810 811 /* Put packet to the user space buffer */ 812 static ssize_t macvtap_put_user(struct macvtap_queue *q, 813 const struct sk_buff *skb, 814 struct iov_iter *iter) 815 { 816 int ret; 817 int vnet_hdr_len = 0; 818 int vlan_offset = 0; 819 int total; 820 821 if (q->flags & IFF_VNET_HDR) { 822 struct virtio_net_hdr vnet_hdr; 823 vnet_hdr_len = q->vnet_hdr_sz; 824 if (iov_iter_count(iter) < vnet_hdr_len) 825 return -EINVAL; 826 827 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, 828 macvtap_is_little_endian(q))) 829 BUG(); 830 831 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 832 sizeof(vnet_hdr)) 833 return -EFAULT; 834 835 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 836 } 837 total = vnet_hdr_len; 838 total += skb->len; 839 840 if (skb_vlan_tag_present(skb)) { 841 struct { 842 __be16 h_vlan_proto; 843 __be16 h_vlan_TCI; 844 } veth; 845 veth.h_vlan_proto = skb->vlan_proto; 846 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 847 848 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 849 total += VLAN_HLEN; 850 851 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 852 if (ret || !iov_iter_count(iter)) 853 goto done; 854 855 ret = copy_to_iter(&veth, sizeof(veth), iter); 856 if (ret != sizeof(veth) || !iov_iter_count(iter)) 857 goto done; 858 } 859 860 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 861 skb->len - vlan_offset); 862 863 done: 864 return ret ? ret : total; 865 } 866 867 static ssize_t macvtap_do_read(struct macvtap_queue *q, 868 struct iov_iter *to, 869 int noblock) 870 { 871 DEFINE_WAIT(wait); 872 struct sk_buff *skb; 873 ssize_t ret = 0; 874 875 if (!iov_iter_count(to)) 876 return 0; 877 878 while (1) { 879 if (!noblock) 880 prepare_to_wait(sk_sleep(&q->sk), &wait, 881 TASK_INTERRUPTIBLE); 882 883 /* Read frames from the queue */ 884 skb = skb_array_consume(&q->skb_array); 885 if (skb) 886 break; 887 if (noblock) { 888 ret = -EAGAIN; 889 break; 890 } 891 if (signal_pending(current)) { 892 ret = -ERESTARTSYS; 893 break; 894 } 895 /* Nothing to read, let's sleep */ 896 schedule(); 897 } 898 if (!noblock) 899 finish_wait(sk_sleep(&q->sk), &wait); 900 901 if (skb) { 902 ret = macvtap_put_user(q, skb, to); 903 if (unlikely(ret < 0)) 904 kfree_skb(skb); 905 else 906 consume_skb(skb); 907 } 908 return ret; 909 } 910 911 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 912 { 913 struct file *file = iocb->ki_filp; 914 struct macvtap_queue *q = file->private_data; 915 ssize_t len = iov_iter_count(to), ret; 916 917 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 918 ret = min_t(ssize_t, ret, len); 919 if (ret > 0) 920 iocb->ki_pos = ret; 921 return ret; 922 } 923 924 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 925 { 926 struct macvlan_dev *vlan; 927 928 ASSERT_RTNL(); 929 vlan = rtnl_dereference(q->vlan); 930 if (vlan) 931 dev_hold(vlan->dev); 932 933 return vlan; 934 } 935 936 static void macvtap_put_vlan(struct macvlan_dev *vlan) 937 { 938 dev_put(vlan->dev); 939 } 940 941 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 942 { 943 struct macvtap_queue *q = file->private_data; 944 struct macvlan_dev *vlan; 945 int ret; 946 947 vlan = macvtap_get_vlan(q); 948 if (!vlan) 949 return -EINVAL; 950 951 if (flags & IFF_ATTACH_QUEUE) 952 ret = macvtap_enable_queue(vlan->dev, file, q); 953 else if (flags & IFF_DETACH_QUEUE) 954 ret = macvtap_disable_queue(q); 955 else 956 ret = -EINVAL; 957 958 macvtap_put_vlan(vlan); 959 return ret; 960 } 961 962 static int set_offload(struct macvtap_queue *q, unsigned long arg) 963 { 964 struct macvlan_dev *vlan; 965 netdev_features_t features; 966 netdev_features_t feature_mask = 0; 967 968 vlan = rtnl_dereference(q->vlan); 969 if (!vlan) 970 return -ENOLINK; 971 972 features = vlan->dev->features; 973 974 if (arg & TUN_F_CSUM) { 975 feature_mask = NETIF_F_HW_CSUM; 976 977 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 978 if (arg & TUN_F_TSO_ECN) 979 feature_mask |= NETIF_F_TSO_ECN; 980 if (arg & TUN_F_TSO4) 981 feature_mask |= NETIF_F_TSO; 982 if (arg & TUN_F_TSO6) 983 feature_mask |= NETIF_F_TSO6; 984 } 985 986 if (arg & TUN_F_UFO) 987 feature_mask |= NETIF_F_UFO; 988 } 989 990 /* tun/tap driver inverts the usage for TSO offloads, where 991 * setting the TSO bit means that the userspace wants to 992 * accept TSO frames and turning it off means that user space 993 * does not support TSO. 994 * For macvtap, we have to invert it to mean the same thing. 995 * When user space turns off TSO, we turn off GSO/LRO so that 996 * user-space will not receive TSO frames. 997 */ 998 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 999 features |= RX_OFFLOADS; 1000 else 1001 features &= ~RX_OFFLOADS; 1002 1003 /* tap_features are the same as features on tun/tap and 1004 * reflect user expectations. 1005 */ 1006 vlan->tap_features = feature_mask; 1007 vlan->set_features = features; 1008 netdev_update_features(vlan->dev); 1009 1010 return 0; 1011 } 1012 1013 /* 1014 * provide compatibility with generic tun/tap interface 1015 */ 1016 static long macvtap_ioctl(struct file *file, unsigned int cmd, 1017 unsigned long arg) 1018 { 1019 struct macvtap_queue *q = file->private_data; 1020 struct macvlan_dev *vlan; 1021 void __user *argp = (void __user *)arg; 1022 struct ifreq __user *ifr = argp; 1023 unsigned int __user *up = argp; 1024 unsigned short u; 1025 int __user *sp = argp; 1026 struct sockaddr sa; 1027 int s; 1028 int ret; 1029 1030 switch (cmd) { 1031 case TUNSETIFF: 1032 /* ignore the name, just look at flags */ 1033 if (get_user(u, &ifr->ifr_flags)) 1034 return -EFAULT; 1035 1036 ret = 0; 1037 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1038 ret = -EINVAL; 1039 else 1040 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1041 1042 return ret; 1043 1044 case TUNGETIFF: 1045 rtnl_lock(); 1046 vlan = macvtap_get_vlan(q); 1047 if (!vlan) { 1048 rtnl_unlock(); 1049 return -ENOLINK; 1050 } 1051 1052 ret = 0; 1053 u = q->flags; 1054 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1055 put_user(u, &ifr->ifr_flags)) 1056 ret = -EFAULT; 1057 macvtap_put_vlan(vlan); 1058 rtnl_unlock(); 1059 return ret; 1060 1061 case TUNSETQUEUE: 1062 if (get_user(u, &ifr->ifr_flags)) 1063 return -EFAULT; 1064 rtnl_lock(); 1065 ret = macvtap_ioctl_set_queue(file, u); 1066 rtnl_unlock(); 1067 return ret; 1068 1069 case TUNGETFEATURES: 1070 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1071 return -EFAULT; 1072 return 0; 1073 1074 case TUNSETSNDBUF: 1075 if (get_user(s, sp)) 1076 return -EFAULT; 1077 1078 q->sk.sk_sndbuf = s; 1079 return 0; 1080 1081 case TUNGETVNETHDRSZ: 1082 s = q->vnet_hdr_sz; 1083 if (put_user(s, sp)) 1084 return -EFAULT; 1085 return 0; 1086 1087 case TUNSETVNETHDRSZ: 1088 if (get_user(s, sp)) 1089 return -EFAULT; 1090 if (s < (int)sizeof(struct virtio_net_hdr)) 1091 return -EINVAL; 1092 1093 q->vnet_hdr_sz = s; 1094 return 0; 1095 1096 case TUNGETVNETLE: 1097 s = !!(q->flags & MACVTAP_VNET_LE); 1098 if (put_user(s, sp)) 1099 return -EFAULT; 1100 return 0; 1101 1102 case TUNSETVNETLE: 1103 if (get_user(s, sp)) 1104 return -EFAULT; 1105 if (s) 1106 q->flags |= MACVTAP_VNET_LE; 1107 else 1108 q->flags &= ~MACVTAP_VNET_LE; 1109 return 0; 1110 1111 case TUNGETVNETBE: 1112 return macvtap_get_vnet_be(q, sp); 1113 1114 case TUNSETVNETBE: 1115 return macvtap_set_vnet_be(q, sp); 1116 1117 case TUNSETOFFLOAD: 1118 /* let the user check for future flags */ 1119 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1120 TUN_F_TSO_ECN | TUN_F_UFO)) 1121 return -EINVAL; 1122 1123 rtnl_lock(); 1124 ret = set_offload(q, arg); 1125 rtnl_unlock(); 1126 return ret; 1127 1128 case SIOCGIFHWADDR: 1129 rtnl_lock(); 1130 vlan = macvtap_get_vlan(q); 1131 if (!vlan) { 1132 rtnl_unlock(); 1133 return -ENOLINK; 1134 } 1135 ret = 0; 1136 u = vlan->dev->type; 1137 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1138 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1139 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1140 ret = -EFAULT; 1141 macvtap_put_vlan(vlan); 1142 rtnl_unlock(); 1143 return ret; 1144 1145 case SIOCSIFHWADDR: 1146 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1147 return -EFAULT; 1148 rtnl_lock(); 1149 vlan = macvtap_get_vlan(q); 1150 if (!vlan) { 1151 rtnl_unlock(); 1152 return -ENOLINK; 1153 } 1154 ret = dev_set_mac_address(vlan->dev, &sa); 1155 macvtap_put_vlan(vlan); 1156 rtnl_unlock(); 1157 return ret; 1158 1159 default: 1160 return -EINVAL; 1161 } 1162 } 1163 1164 #ifdef CONFIG_COMPAT 1165 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1166 unsigned long arg) 1167 { 1168 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1169 } 1170 #endif 1171 1172 static const struct file_operations macvtap_fops = { 1173 .owner = THIS_MODULE, 1174 .open = macvtap_open, 1175 .release = macvtap_release, 1176 .read_iter = macvtap_read_iter, 1177 .write_iter = macvtap_write_iter, 1178 .poll = macvtap_poll, 1179 .llseek = no_llseek, 1180 .unlocked_ioctl = macvtap_ioctl, 1181 #ifdef CONFIG_COMPAT 1182 .compat_ioctl = macvtap_compat_ioctl, 1183 #endif 1184 }; 1185 1186 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1187 size_t total_len) 1188 { 1189 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1190 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1191 } 1192 1193 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1194 size_t total_len, int flags) 1195 { 1196 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1197 int ret; 1198 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1199 return -EINVAL; 1200 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1201 if (ret > total_len) { 1202 m->msg_flags |= MSG_TRUNC; 1203 ret = flags & MSG_TRUNC ? ret : total_len; 1204 } 1205 return ret; 1206 } 1207 1208 static int macvtap_peek_len(struct socket *sock) 1209 { 1210 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, 1211 sock); 1212 return skb_array_peek_len(&q->skb_array); 1213 } 1214 1215 /* Ops structure to mimic raw sockets with tun */ 1216 static const struct proto_ops macvtap_socket_ops = { 1217 .sendmsg = macvtap_sendmsg, 1218 .recvmsg = macvtap_recvmsg, 1219 .peek_len = macvtap_peek_len, 1220 }; 1221 1222 /* Get an underlying socket object from tun file. Returns error unless file is 1223 * attached to a device. The returned object works like a packet socket, it 1224 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1225 * holding a reference to the file for as long as the socket is in use. */ 1226 struct socket *macvtap_get_socket(struct file *file) 1227 { 1228 struct macvtap_queue *q; 1229 if (file->f_op != &macvtap_fops) 1230 return ERR_PTR(-EINVAL); 1231 q = file->private_data; 1232 if (!q) 1233 return ERR_PTR(-EBADFD); 1234 return &q->sock; 1235 } 1236 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1237 1238 static int macvtap_queue_resize(struct macvlan_dev *vlan) 1239 { 1240 struct net_device *dev = vlan->dev; 1241 struct macvtap_queue *q; 1242 struct skb_array **arrays; 1243 int n = vlan->numqueues; 1244 int ret, i = 0; 1245 1246 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL); 1247 if (!arrays) 1248 return -ENOMEM; 1249 1250 list_for_each_entry(q, &vlan->queue_list, next) 1251 arrays[i++] = &q->skb_array; 1252 1253 ret = skb_array_resize_multiple(arrays, n, 1254 dev->tx_queue_len, GFP_KERNEL); 1255 1256 kfree(arrays); 1257 return ret; 1258 } 1259 1260 static int macvtap_device_event(struct notifier_block *unused, 1261 unsigned long event, void *ptr) 1262 { 1263 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1264 struct macvlan_dev *vlan; 1265 struct device *classdev; 1266 dev_t devt; 1267 int err; 1268 char tap_name[IFNAMSIZ]; 1269 1270 if (dev->rtnl_link_ops != &macvtap_link_ops) 1271 return NOTIFY_DONE; 1272 1273 snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex); 1274 vlan = netdev_priv(dev); 1275 1276 switch (event) { 1277 case NETDEV_REGISTER: 1278 /* Create the device node here after the network device has 1279 * been registered but before register_netdevice has 1280 * finished running. 1281 */ 1282 err = macvtap_get_minor(vlan); 1283 if (err) 1284 return notifier_from_errno(err); 1285 1286 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1287 classdev = device_create(&macvtap_class, &dev->dev, devt, 1288 dev, tap_name); 1289 if (IS_ERR(classdev)) { 1290 macvtap_free_minor(vlan); 1291 return notifier_from_errno(PTR_ERR(classdev)); 1292 } 1293 err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj, 1294 tap_name); 1295 if (err) 1296 return notifier_from_errno(err); 1297 break; 1298 case NETDEV_UNREGISTER: 1299 /* vlan->minor == 0 if NETDEV_REGISTER above failed */ 1300 if (vlan->minor == 0) 1301 break; 1302 sysfs_remove_link(&dev->dev.kobj, tap_name); 1303 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1304 device_destroy(&macvtap_class, devt); 1305 macvtap_free_minor(vlan); 1306 break; 1307 case NETDEV_CHANGE_TX_QUEUE_LEN: 1308 if (macvtap_queue_resize(vlan)) 1309 return NOTIFY_BAD; 1310 break; 1311 } 1312 1313 return NOTIFY_DONE; 1314 } 1315 1316 static struct notifier_block macvtap_notifier_block __read_mostly = { 1317 .notifier_call = macvtap_device_event, 1318 }; 1319 1320 static int macvtap_init(void) 1321 { 1322 int err; 1323 1324 err = alloc_chrdev_region(&macvtap_major, 0, 1325 MACVTAP_NUM_DEVS, "macvtap"); 1326 if (err) 1327 goto out1; 1328 1329 cdev_init(&macvtap_cdev, &macvtap_fops); 1330 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1331 if (err) 1332 goto out2; 1333 1334 err = class_register(&macvtap_class); 1335 if (err) 1336 goto out3; 1337 1338 err = register_netdevice_notifier(&macvtap_notifier_block); 1339 if (err) 1340 goto out4; 1341 1342 err = macvlan_link_register(&macvtap_link_ops); 1343 if (err) 1344 goto out5; 1345 1346 return 0; 1347 1348 out5: 1349 unregister_netdevice_notifier(&macvtap_notifier_block); 1350 out4: 1351 class_unregister(&macvtap_class); 1352 out3: 1353 cdev_del(&macvtap_cdev); 1354 out2: 1355 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1356 out1: 1357 return err; 1358 } 1359 module_init(macvtap_init); 1360 1361 static void macvtap_exit(void) 1362 { 1363 rtnl_link_unregister(&macvtap_link_ops); 1364 unregister_netdevice_notifier(&macvtap_notifier_block); 1365 class_unregister(&macvtap_class); 1366 cdev_del(&macvtap_cdev); 1367 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1368 idr_destroy(&minor_idr); 1369 } 1370 module_exit(macvtap_exit); 1371 1372 MODULE_ALIAS_RTNL_LINK("macvtap"); 1373 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1374 MODULE_LICENSE("GPL"); 1375